Hi Danny, thanks a lot for your quick response to analyse the problems in the ball package. I take the freedom to turn this discussion into a public one and CC the relevant bugs to leave a record there that something is happening.
On Thu, Nov 10, 2016 at 03:26:30PM +0100, Danny Edel wrote: > On 11/09/2016 03:39 PM, Andreas Tille wrote: > > Sure you can't give a time line. I think when we start our Advent bug > > squashing party we need to tackle it somehow - but having this sorted > > out before to enable some testing would be great. > > I have had some time to look into this, there are three failures, and > all are perfectly reproducible. Investigating (with core dumps) revealed: > > 1. PoseClustering_Test depends on verbatim boost::serialisation output, > which changes with every library version update. Upstream has already > fixed this by ignoring the actual file contents, and instead feeding the > file to the de-serializer and check if the resulting object compares equal. Sounds good. > 2. DefaultProcessors_test segfaults after trying to use an optimized-out > variable. This looks like a legitimate upstream bug to me. > > 3. XDRPersistenceManager_test segfaults after following a null pointer. > Also looks like an upstream bug. > > While backporting the fix for (1) is straightforward, the code paths > responsible for (2) and (3) have seen quite some activity in the > meantime, with the relevant commits changing lots of other files. Hmmm, may be we should base the packaging on a later upstream commit? > Among > other things, upstream also migrated from Qt4 to Qt5, and incorporated a > few fixes for recent boost. We somehow should target to Qt5 anyway (see #784451) better sooner than later. > I tried building upstream master to see if it still contained the bugs. > It did not, however one other test fails > (AssignBondOrderProcessor_Test), which is related to upstream issue 576. > > I will try to isolate and backport fixes for the three issues, and > report back afterwards. Again thanks a lot for your contribution Andreas. -- http://fam-tille.de